Processor cluster migration techniques

• US 9,471,395 B2
• Filed: 08/23/2012
• Issued: 10/18/2016
• Est. Priority Date: 08/23/2012
• Status: Active Grant

• Current Assignee: NVIDIA Corporation
• Original Assignee: NVIDIA Corporation
• Inventors: Ahmad, Sagheer, Chaudhry, Shailender, Mathieson, John George, Overby, Mark Alan
• Primary Examiner(s): CAO, DIEM K
• Application Number:
  US13/592,748
• Publication Number:
  US 20140059548A1
• Time in Patent Office: 1,517 Days
• Field of Search: 713/320, 713/324, 718/713
• US Class Current: 1/1
• CPC Class Codes: G06F 2209/5019 Workload prediction
• CPC Class Codes: G06F 2209/5022 Workload threshold
• CPC Class Codes: G06F 9/5088 involving task migration
• CPC Class Codes: G06F 9/5094 where the allocation takes into account power or heat criteria power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

Embodiments of the present technology provide for migrating processes executing one any one of a plurality of cores in a multi-core cluster to a core of a separate cluster without first having to transfer the processes to a predetermined core of the multi-core cluster. Similarly, the processes may be transferred from the core of the separate cluster to the given core of the multi-core cluster.

Claims

1. A method comprising;

power-ungating a core of a single-core cluster, when one or more processes are executing on a given one of any core of the multi-core cluster and utilization of the given core is predicted to be in a first predetermined range, wherein processing performance of each core of the multi-core cluster is greater than the core of the single core cluster;

saving a context of the one or more processes executing on the given core of the multi-core cluster when the one or more processes are executing on the given core of the multi-core cluster and the utilization of the given core of the multi-core cluster is predicted to be in the first predetermined range;

re-mapping one or more shared resources utilized by the given core of the multi-core cluster to the core of the single-core cluster, including emulating the identity of the given core of the multi-core cluster by the core of the single core-cluster, after power-ungating the core of the single-core cluster and saving, the context of the one or more processes executing on the given core of the multi-core cluster, wherein the shared resources are common to the multi-core cluster and the single-core cluster, and the multi-core cluster, single-core cluster and shared resources are fabricated together on an integrated circuit die;

restarting execution of the saved context on the core of the single-core cluster after re-mapping the one or more ;

hared resources of the given core of the multi-core cluster to the core of the single-core cluster; and

power-gating the cores of the multi-core cluster after saving the context of the one or more processes executing on the given core of the multi-core cluster.

2. The method of claim 1, wherein the context of the one or more processes executing on the given core of the multi-core cluster is stored and the one or more shared resources utilized by the given core of the multi-core cluster are re-mapped to the core of the single-core cluster during an idle period of the one or more processes executing on the given core of the multi-core cluster.

3. The method of claim 1, wherein saving the context of the one or more processes executing on the given core of the multi-core cluster, re-mapping the one or more shared resources utilized by the given core of the multi-core cluster to the core of the single-core cluster, and restarting execution of the saved context on the core of the single-core cluster is performed underneath an operating system.

4. The method of claim 1, wherein utilization of the given core is predicted from past execution history on the given core of the multi-core cluster.

5. The method of claim 1, wherein utilization of the given core is predicted from pending scheduling on the given core of the multi-core cluster.

6. The method of claim 1, wherein utilization of the given core is predicted from past execution history and pending scheduling on the given core of the multi-core cluster.

7. The method of claim 1, wherein re-mapping one or more shared resources utilized by the given core of the multi-core cluster to the core of the single-core cluster comprises changing an identifier of the core of the single-core cluster to an identifier of the given core of the multi-core cluster.

8. The method of claim 7, wherein re-mapping the one or more shared resources utilized by the given core of the multi-core cluster to the core of the single-core cluster comprises multiplexing the one or more shared resourced to the core of the single-core cluster based upon the changed identifier of the single-core cluster.

9. The method of claim 1, wherein the one or more shared resources are selected from a group consisting of one or more timers, and one or more interrupts.

10. The method of claim 1, further comprising:

holding off a request to start one or more other processes on one or more other cores of the multi-core cluster, when one or more processes are executing on the core of the single-core cluster;

power-ungating the given core and the one or more other cores of the multi-core cluster in response to the request to start the one or more other processes on the one or more other cores of the multi-core cluster;

saving the context of the one or more processes executing on the core of the single-core cluster in response to the request to start the one or more other processes on the one or more other cores of the multi-core cluster;

un-mapping the one or more shared resources utilized by the given core of the multi-core cluster from the core of the single-core cluster alter power-ungating the given core of themulti-core cluster and saving the context of the one or more processes executing on the core of the single-core cluster;

restarting execution of the saved context, of the one or more processes that were executing on the core of the single-core cluster, on the given core of the multi-core cluster after un-mapping the one or more shared resources utilized by the given core of the multi-core cluster from the core of the single-core cluster;

starting execution of the one or more other processes on the one or more other cores of the multi-core cluster after un-mapping the one or more shared resources of the given core of the multi-core cluster from the core of the single-core cluster; and

power-gating the core of the single-core cluster after saving the context of the one or more process executing on the core of the single-core cluster.

11. The method of claim 10, wherein holding off a request to start the one or more other processes on one or more other cores of the multi-core cluster comprises holding off an inter-process interrupt (IPI) requesting to start the one or more other processes on the one or more other of the cores of the multi-core cluster.

12. The method of claim 10, wherein restarting execution of the saved context, of the one or more processes that were executing on the core of the single-core cluster, on the given core of the multi-core cluster comprises executing a reset vector pointing to the saved context by the given core of the multi-core cluster.

13. The method of claim 1, further comprising:

power-ungating the given core of the multi-core cluster when utilization of the core of the single core cluster is predicted to be in a second predetermined range;

saving the context of the one or more processes executing on the core of the single-core cluster when utilization of the core of the single core cluster is predicted to be in the second predetermined range;

un-mapping the one or more resources utilized by the given core of the multi-core cluster from the core of the single-core cluster after power-ungating the given core of the multi-core cluster and saving the context of the one or more processes executing on the core of the single-core cluster;

restarting execution of the saved context, of the one or more processes that were executing on the core of the single-core cluster, on the given core of the multi-core cluster after un-mapping the one or more resources of the given core of the multi-core cluster from the core of the single-core cluster; and

power-gating the core of the single-core cluster after saving the context of the one or more processes executing, on the core of the single-core cluster.

14. An apparatus comprising:

a means for determining if processes are executing on a given one of any core of a multi-core cluster and that the Utilization of the given tore is predicted to be in a first predetermined range;

a means for power-on gating a core of a single-core cluster, when one or more processes are executing on the given core of the multi-core cluster and utilization of the given core is predicted to be in the first predetermined range, wherein processing performance of each core of the multi-core cluster is greater than the core of the single core cluster;

a means for saving a context of the one or more processes executing on the given core of the multi-core cluster when one or more processes are executing on the given core of the multi- core cluster and the utilization of the given core of the multi-core cluster is predicted to be in a predetermined range;

a means for cluster switching one or more shared resources utilized by the given core of the multi-core cluster to the core of the single-core cluster, including emulating the identity of the given core of the multi-core cluster by the core of the single core-cluster, after power-ungating the core of the single-core cluster and saving the context of the one or more processes executing on the given core of the multi-core cluster, wherein the shared resources are common to the multi-core cluster and the single-core cluster, wherein the multi-core cluster, single-core cluster and shared resources are fabricated together on a system-on-a-chip (SoC);

a means for restarting execution of the saved context on the core or the single-core cluster after re-mapping the one or more shared resources of the given core of the multi-core cluster to the core of the single-core cluster; and

a means for power-off gating the cores of the multi-core cluster after saving the context of the one or more processes executing on the given core of the multi-core cluster.

15. The apparatus according to claim 14, farther comprising:

a means for predicting that the utilization of the core of the single-core cluster is within a second predetermined range;

a means for power-ungating the given core of the multi-core cluster when utilization of the core of the single core cluster is predicted to he in a second predetermined range;

means for saving the context of the one or more processes executing on the core of the single-core cluster when utilization of the core of the single core cluster is predicted to be in the second predetermined range;

a means for cluster switching the one or more shared resources utilized by the given core of the multi-core cluster from the core of the single-core cluster after power-ungating the given core of the multi-core cluster and saving the context of the one or more processes executing on the core of the single-core cluster;

a means for restarting execution of the saved context on the given core of the multi-core cluster after cluster switching the one or more shared resources to the given core of the multi-core cluster from the core of the single-core cluster; and

a means for power-gating the core of the single-core cluster after saving the context of the one or more processes executing on the core of the single-core cluster.

16. The apparatus according to claim 14, further comprising:

a means for holding off a request to start one or more other processes on one or more other cores of the multi-core cluster when one or more processes are executing on the core of the single-core cluster;

a means for power-ungating the given core and one or more other cores of the multi-core cluster in response to the request to start one or more other processes on the one or more other cores of the multi-core cluster;

a means for saving the context of the one or more process executing on the core of the single-core cluster in response to the request to start one or more other processes on the one or more other cores of the multi-core cluster;

a means for cluster switching the one or more shared resources utilized by the given core of the multi-core cluster from the core of the single-core cluster after power-ungating the given core of the multi-core cluster and saving the context of the one or more processes executing on the core of the single-core cluster;

a means for restarting execution of the saved context on the given core of the multi-core cluster after un-mapping one or more resources of the given core of the multi-core cluster from the core of the single-core cluster;

a means for starting, execution of the one or more other processes on the one or more other cores of the multi-core cluster after cluster switching the one or more shared resources to the given core of the multi-core cluster from the core of the single-core cluster; and

a means for power-gating the core of the single-core cluster after saving the context of the one or more processes executing on the core of the single-core cluster.

17. The system according to claim 14, wherein the means of power gating, saving contexts, cluster switching and restarting execution of saved contexts does not utilize software virtualization.

18. The system according to claim 14, wherein the means for cluster switching the one or more shared resources utilized by the given core of the multi-core cluster to the core of the single-core cluster is based upon an identifier of the given core of the multi-core cluster.

19. The system according to claim 14, a means for changing an identifier of the single core of the single-core cluster to the identifier of the given core of the multi-core cluster.

20. The system according to claim 14, wherein the one or more shared resources are selected from a group consisting of one or more timers, and one or more interrupts.